第3-1章：使用 LangChain 通过LLM查询DataFrames-大模型应用实践课程

!pip install -q chromadb==0.4.22
!pip install -q langchain==0.1.4
!pip install -q sentence_transformers==2.3.0

from google.colab import drive
drive.mount('/content/drive')
Mounted at /content/drive
!pip install -q kaggle
import os
#This directory should contain you kaggle.json file with your key
os.environ['KAGGLE_CONFIG_DIR'] = '/content/drive/MyDrive/kaggle'
!kaggle datasets download -d kotartemiy/topic-labeled-news-dataset
#!kaggle datasets download -d gpreda/bbc-news
Dataset URL: https://www.kaggle.com/datasets/kotartemiy/topic-labeled-news-dataset
License(s): CC0-1.0
Downloading topic-labeled-news-dataset.zip to /content
 85% 8.00M/9.45M [00:01<00:00, 12.7MB/s]
100% 9.45M/9.45M [00:01<00:00, 8.75MB/s]
import numpy as np
import pandas as pd
import zipfile

# Define the path to your zip file
file_path = '/content/topic-labeled-news-dataset.zip'
#file_path = '/content/bbc-news.zip'
with zipfile.ZipFile(file_path, 'r') as zip_ref:
    zip_ref.extractall('/content/drive/MyDrive/kaggle')
news = pd.read_csv('/content/drive/MyDrive/kaggle/labelled_newscatcher_dataset.csv', sep=';')
MAX_NEWS = 1000
DOCUMENT="title"
TOPIC="topic"

#news = pd.read_csv('/content/drive/MyDrive/kaggle/bbc_news.csv')
#MAX_NEWS = 500
#DOCUMENT="description"
#TOPIC="title"

#Because it is just a course we select a small portion of News.
subset_news = news.head(MAX_NEWS)

news.head(2)

from langchain.document_loaders import DataFrameLoader
from langchain.vectorstores import Chroma

df_loader = DataFrameLoader(subset_news, page_content_column=DOCUMENT)
## Then, we use the loader to load the document.

df_document = df_loader.load()
display(df_document[:2])

from langchain.text_splitter import CharacterTextSplitter
#from langchain.embeddings import HuggingFaceEmbeddings

text_splitter = CharacterTextSplitter(chunk_size=250, chunk_overlap=10)
texts = text_splitter.split_documents(df_document)
display(texts[:2])

from langchain.embeddings.sentence_transformer import SentenceTransformerEmbeddings
embedding_function = SentenceTransformerEmbeddings(model_name="all-MiniLM-L6-v2")

directory_cdb = '/content/drive/MyDrive/chromadb'
chroma_db = Chroma.from_documents(
    texts, embedding_function, persist_directory=directory_cdb
)

from langchain.chains import RetrievalQA
from langchain.llms import HuggingFacePipeline
#from transformers import pipeline

retriever = chroma_db.as_retriever()

model_id = "databricks/dolly-v2-3b" #a good textgeneration model for testing
task="text-generation"

#model_id = "google/flan-t5-large" #Nice text2text model
#task="text2text-generation"

#If you want to test the  T5 model remove the return_full_text parameter in pipeline_kwargs,
#also I recommend to add model_kwargs and change the temperature.
#    model_kwargs={
#        "temperature": 0,
#        "max_length": 256
#    },
hf_llm = HuggingFacePipeline.from_model_id(
    model_id=model_id,
    task=task,
    pipeline_kwargs={
        "max_new_tokens": 256,
        "repetition_penalty":1.1,
        "return_full_text":True
    },
)

document_qa = RetrievalQA.from_chain_type(
    llm=hf_llm, retriever=retriever, chain_type='stuff'
)

#Sample question for newscatcher dataset.
response = document_qa.invoke("Can I buy a Toshiba laptop?")

#Sample question for BBC Dataset.
#response = document_qa.run("Who is going to meet boris johnson?")
display(response)

from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough
from langchain_core.output_parsers import StrOutputParser

template = """Answer the question based on the following context:
{context}

Question: {question}
"""
prompt = ChatPromptTemplate.from_template(template)
chain = (
    {"context": retriever, "question": RunnablePassthrough()}
    | prompt
    | hf_llm
    | StrOutputParser()
)
chain.invoke("Can I buy a Toshiba laptop?")

embedding_s1 = embedding_function.embed_query(
    "I would like to eat more vegetables and exercise every day")

embedding_s2 = embedding_function.embed_query(
    "I will try to maintain a healthier lifestyle.")

embedding_s3 = embedding_function.embed_query(
    "I prefer to play football")

print(f""" embedding_s1 = {len(embedding_s1)}
 embedding_s2 =  {len(embedding_s2)}
 embedding_s3 =  {len(embedding_s3)}""")

# 5 first positions of embedding_s1.
print(embedding_s1[:5])
print(embedding_s2[:5])
print(embedding_s3[:5])

!pip install -q scikit-learn==1.2.2
from sklearn.metrics.pairwise import cosine_similarity
embedding_s1_2d = np.array(embedding_s1).reshape(1, -1)
embedding_s2_2d = np.array(embedding_s2).reshape(1, -1)
embedding_s3_2d = np.array(embedding_s3).reshape(1, -1)
print(embedding_s1_2d[0][:5])

print(cosine_similarity(embedding_s1_2d, embedding_s2_2d))
print(cosine_similarity(embedding_s1_2d, embedding_s3_2d))
print(cosine_similarity(embedding_s2_2d, embedding_s3_2d))

import matplotlib.pyplot as plt
from sklearn.decomposition import PCA

# Perform PCA for 2D visualization
PCA_model = PCA(n_components = 2)
embeddings_sentences=[]
embeddings_sentences.append(embedding_s1)
embeddings_sentences.append(embedding_s2)
embeddings_sentences.append(embedding_s3)
PCA_model.fit(embeddings_sentences)
embeddings_coord = PCA_model.transform(embeddings_sentences)
print(embeddings_coord)

import matplotlib.pyplot as plt

x, y = zip(*embeddings_coord)

# Name the points
names = ['s1', 's2', 's3']
# Colors
colors = ['red', 'green', 'blue']

for i, name in enumerate(names):
    plt.scatter(x[i], y[i], marker='o', color=colors[i], label=name)

plt.xlabel('X')
plt.ylabel('Y')
plt.legend()
plt.show()